Speaker: Prof. Kuang-Hao Liu
(Institute of Commucations Engineering, National Tsing Hua University)
The vast availability of Millimeter wave (mmWave) spectrum holds the promise of data rates several tens of times faster than those of mid- and low-bands. However, a significant challenge of mmWave communication lies in the severe propagation loss, necessitating compensation through directional beamforming. Accurate beamforming becomes particularly challenging in scenarios involving user movement. This talk begins by elucidating the fundamental principle of beam alignment, outlining its importance in mmWave communication systems. Subsequently, it introduces a novel approach that harnesses the predictive capability of machine learning (ML) models to achieve high beam alignment accuracy. Through this innovative method, the talk aims to address the challenges posed by user mobility in mmWave communication systems.
Speaker: Prof. Cheng-Hsin Hsu
(National Tsing Hua University, Department of Computer Science)
Cloud VR gaming shifts computationally intensive games to cloud servers, alleviating strain on Head-Mounted Displays. Yet, ensuring high-quality experiences faces challenges like human perception complexities and network variability. We introduce dynamic foveation techniques to enhance gamer Quality of Experience (QoE) by prioritizing resources to foveal regions, improving visual fidelity and bandwidth usage. We next present predictive models to adjust encoding settings for better gamer QoE. Our studies stream 2D video codecs from cloud servers to HMD clients, with the potential for emerging 3D representations. Along that direction, we share a novel error concealment pipeline for dynamic 3D point clouds, enhancing streaming quality under high packet loss rates. These advancements in cloud VR gaming open avenues for future research and development.
Speaker: Prof. Min-Chun Hu
(Department of Computer Science, National Tsing Hua University)
Film artworks, with their rich blend of visuals and soundtracks, adeptly translate the emotional visions of directors and creators, making them ideal stimuli in numerous studies aimed at evoking emotions. This talk introduces an approach to creating an affective common embedding space for music and videos, inspired by the essence of film artworks. This method not only bridges the heterogeneity gap between these mediums but also recommends suitable music for videos and vice versa, thereby amplifying the emotional resonance in multimedia. Additionally, we have crafted an interactive game designed to make the labeling of affective states more engaging. The game fosters interactions between participants, forming a dynamic loop.
Speaker: A.Prof. Ping-Hsuan Han
(National Taipei University of Technology )
Due to the growing popularity of virtual reality (VR) and augmented reality (AR) head-mounted displays, haptic technology has become a prevalent research topic again. Furthermore, numerous unprecedented applications, technologies, and experiences were aroused in this field. Emerging experiences usually accompany the development of emerging technologies. However, emerging experiences are not necessarily produced by emerging technologies. This speech will share with you (1)multisensory technologies and (2)immersive experience design based on our recent international research and publications as well as domestic industry-academia cooperation cases, which include the research in HCI, the method of interactive design and the technique of dissemination and demonstration.
Speaker: Prof. Chia-Wen Lin
(Department of Electrical Engineering, National Tsing Hua University)
As a new sensing technology, Terahertz (THz) computational imaging has recently attracted significant attention thanks to its non-invasive, non-destructive, non-ionizing, material-classification, and ultra-fast nature for 3D object exploration and inspection. However, its strong water absorption nature and low noise tolerance lead to undesired blurs and distortions of reconstructed THz images. The performances of existing methods are highly constrained by the diffraction-limited THz signals. In this talk, we will introduce the characteristics of THz imaging and its applications. We will also show how to break the limitations of THz imaging with the aid of learned complementary information between the THz amplitude and phase images sampled at prominent frequencies (i.e., the water absorption profile of THz signal) for THz image restoration.
Speaker: Prof. Chao-Tsung Huang
(Department of Electrical Engineering, National Tsing Hua University)
In the era of artificial intelligence, convolutional neural networks (CNNs) are emerging as a powerful technique for image processing, such as denoising, super-resolution, and even style transfer. They show great potential to bring next-generation cameras and displays to our daily life. However, it is difficult for conventional accelerators to generate ultra-high-resolution videos at the edge due to considerable DRAM bandwidth and power consumption. In this talk, I will introduce two of our recent works on tackling these challenges. The first work—CINE—is a computational imaging neural engine with overlapped stripe inference flow and structure-sparse convolution. Given only 0.9GB/s of DRAM bandwidth, it provides 4K-UHD applications with 4.6-8.3 TOPS/W of efficiency. The second work—VISTA—achieves video/image spatial/temporal interpolation acceleration with cuboid-based layer fusion and ring-algebra computation sparsity. It supports video CNNs for 4K-UHD displays while consuming only 704 mW of power and less than 3 GB/s of DRAM bandwidth.
Speaker: Prof.Shervin Shirmohammadi
(the school of Electrial Enginnering and Computer Science, University of Ottawa, Canada)
Title: Towards Superintelligent Network Operation Centers
Abstract: In this talk, I will present some highlights from my research projects in collaboration with Canadian company Ciena Corp. on the use of machine learning (ML) for automating network operation centers (NOC). In today’s NOCs, non-trivial issues are resolved manually and with the collective human intelligence through specialized support teams organized by technology domains, who themselves communicate with the equipment vendors’ technical support. In this talk, we will see the application of ML to automate those operations. Specifically, we will see how ML can be used to diagnose and localize network faults, and to recommend actions to remedy faults and/or improve performance. Of particular interest will be the application of reinforcement learning to manage an NOC with a performance that’s better than human-defined expert rules, potentially leading to the superintelligent management of a network.
Speaker: Prof.Chi Chun Lee
(Department of Electrical Engineering, National Tsing Hua University)
This talk will explore trustworthy AI in speech and emotion recognition. With the market and applications in various fields growing rapidly, researches concerning its basic knowledge as well as advanced techniques has also become more important. Recently, the focuses of AI study has evolved from data itself to users providing/accessing the data -- and this is where trustworthy starts to matter. In this talk, we will introduce several works of our own in tackling the key challenges of trustworthy AI: safety, privacy, and fairness. We hope to continuously develop human-centered technology that enhances usability while balancing ethical considerations in the AI eco-systems.
Speaker: Prof. Chuang-Wen You
(Graduate Institute of Art and Technology, National Tsing Hua University)
Drug addiction is a chronic condition, marked by compulsive drug use. In previous research, cue exposure and biofeedback technologies proved effective in drug psychotherapy sessions; however, the focus has generally been on the awareness of cravings and the identification of cues. There has been relatively little research on methods aimed at facilitating therapist-patient communication, particularly from a user-centered perspective. In this paper, we describe a qualitative technology probe study exploring the means by which patients identify cues and perceive cravings as well as the way that they communicate with therapists. Our analysis considers the difficulties in cue identification and craving perception, the interactions between the two, and the means by which these characteristics could impact the design of VR support systems in the future.
Speaker: Prof. Jen-Yuan Chang
(Department of Mechanical Engineering, National Tsing Hua University)
To contend with an aging population and limitation of availability of healthcare manpower and human operators in manufacturing, the adaptation of robotics in rehabilitation as well as to assist human operations has increased, in particular in developed countries where aging is playing significant role in resulting in long-term disabilities. Strokes, which are primary contributed by hypotonia and chronic hemiparesis, have been found to be one of the major causes for the long-term disabilities, leading to limb/hand functional impairments in patients. Due to limitation of muscle motor capabilities, hyperexcitability of the stretch reflex is commonly found in stroke patients. The so-called “stiffness” or “tightness” of muscles in the stroke patients is referred to the muscle spasticity which is caused by hyperexcitability of motoneurons. In medical practices, it is found that with precise and repeated range of motion (ROM) exercises, the aforementioned flexor hyperexcitability can be reduced. In present practices, the ROM exercises are operated by therapist’s hands to temporarily reduce the severity of spasticity. In evidence-based medicine, it is demonstrated that with robot-assisted rehabilitation, the high-repetitive and high-precision movements can greatly improve the quality of rehabilitation for stroke patients. In this keynote speech, viewing the robotic hand devices as smart machines, Professor Chang will first discuss government’s strategic aims in long-term care and digital precision medicine as well as the global market trend in rehabilitation robots. Then the robotic technologies suitable for rehabilitation, in particular the mechanism design considering biomechanics, will be addressed by sharing his R&D experience in the development of wearable robotic hand/finger devices to assistive rehabilitation, of which devices had obtained Taiwan and US FDA approvals and ISO13485/GMP certifications. Last but not the least, strategic approaches in transforming technologies into commercial products will be discussed in the context of combining university R&D, government resource as well as venture capital opportunities. Prof. Chang will also share with the audience on his recently R&D on dual-arm humanoid robot named “Tsinghua Gentleman Robot” that possesses anthropomorphic hands and certain intelligence which is able to perform human-like operations.
Speaker: Prof. Zong-Hong Lin
(Department of Bio-engineering, National Taiwan University)
Designing devices with self-powered sensing function has become a popular research field since its emergence in recent decades. Triboelectrification occurs when two materials come into contact with each other, causing charge transfer that leads to oppositely charged surfaces; the amount of charge transfer varies depending on material composition.By combining triboelectrification with electrostatic induction, relevant devices can be designed. If used for energy collection purposes, scientists generally refer to them as Triboelectric Nanogenerators (TENGs); but if used for self-powered sensing purposes, scientists call them Triboelectric Nanosensors (TENSs).In the past few years, we have further developed solid-liquid TENSs for measuring targets such as metal ions, small molecules, proteins and microorganisms. Compared with our previously developed solid-solid TENSs, we have not only improved several shortcomings but also established sensing mechanism and working principle which are very important research achievements in this field.
